Accounts payable automation: what good looks like in 2026

Accounts payable is one of the oldest automation targets in business software, which is exactly why so many finance teams are sceptical of the latest wave of it. Here is what has actually changed, what the manual version costs you today, and what to automate first.

What does accounts payable automation actually do?

An invoice arrives by email or post. The system reads it, extracts the supplier, amount, line items and due date, matches it against a purchase order and delivery note where one exists, flags anything that does not match, and posts the clean ones to your accounting software with the original document attached. A person reviews the exceptions and approves payment runs. Nobody re-types a number from a PDF into a ledger.

That description has been true of AP automation software for over a decade. What changed with AI is the reading step: older systems used fixed templates and struggled with a new supplier's invoice layout; a language model reads an unfamiliar invoice roughly as well as a person does, so the software handles the messy long tail of one-off suppliers and odd formats that used to fall back to manual entry.

How much does manual invoice processing really cost?

The best published benchmarks are American, since nobody runs a comparable UK-wide survey, but the shape translates directly. Ardent Partners' State of ePayables research puts the average fully loaded cost of processing one invoice by hand at around $10.89, against about $2.78 at organisations it rates best-in-class on automation, a roughly 75% gap. The same research puts average processing time at 10.9 days per invoice against 3.1 days for the best performers, and found 73% of accounts payable departments now use some form of automation, up from 56% in 2022. Convert those figures loosely to pounds and multiply by however many invoices your business processes a month, and the annual gap between "average" and "automated" is rarely small once you add it up.

What should you automate first?

• Data capture and matching. Reading the invoice and checking it against a purchase order is the highest-volume, most rule-bound step, and the one with the clearest right answer most of the time. It is also where the £-per-hour saving is largest, because it is the step a person currently does by hand for every single invoice.
• Routing and chasing. Getting an invoice to the right approver and following up when it stalls is pure plumbing. Automating it removes the "which pile is this invoice in" problem without touching anyone's judgement.
• Posting to the ledger. Once an invoice is matched and approved, posting it is a mechanical step worth removing entirely.

What should stay with a person?

Anything that decides whether money actually leaves the business. Invoice and mandate fraud, where a criminal impersonates a supplier or changes bank details to redirect a payment, is falling as a share of UK payment fraud: UK Finance's Annual Fraud Report 2026 puts total UK payment fraud losses at £1.28 billion for 2025, with invoice and mandate-style scams now under a quarter of authorised push payment losses, down from over half in 2020. That fall is partly because banks and businesses got better at exactly the control that matters here: a human checking anything unusual before a payment goes out, especially a changed bank detail or a new supplier. Automating the reading and matching of an invoice is safe. Automating the final release of payment on anything flagged, unfamiliar or above a threshold is not, and it is the one step M22 would not build without a person in the loop (see why human-in-the-loop matters).

Is this worth doing if you only process a few hundred invoices a month?

Price it the same way you would price any automation: hours currently spent, at what it actually costs to employ the person spending them, against the cost of building and running the fix. A business processing 300 invoices a month at, say, 15 minutes each of manual handling is spending roughly 75 hours a month on data entry alone, before chasing and exceptions. Whether that is worth automating depends on what those 75 hours cost you and what a system to remove most of them would cost to build and run, which is exactly the sum a proper AI audit should do before you commit to anything. For the same comparison applied to a single role rather than a process, see AI agent cost vs salary.

The ONS's most recent survey found AI adoption among UK businesses still rising through early 2026, but adoption is not the same as a working system: reading about invoice automation and having invoices flowing through one automatically are different projects, and the gap between them is where most of the value, and most of the risk, actually sits. Start with the highest-volume, most rule-bound step (matching invoices to purchase orders), prove it works on a month of real invoices before trusting it on all of them, and keep a person on anything that authorises money to move. That order rarely goes wrong; the reverse order, automating payment release before the reading and matching has been proven, is how these projects lose trust.

How does M22 help with this?

M22 Consultancy runs the AI Audit first: M22 follows your actual invoices through the business, prices the hours spent on data entry, chasing and exceptions, and ranks what is worth automating before anything gets built. Where matching and posting are safe to automate, M22 builds a workflow automation you own outright; where money is about to leave the business, M22 keeps a person approving the payment, as this article sets out. The AI Audit starts at £1,500, a fixed fee agreed before day one. Book a thirty-minute call to talk through your invoice process, or see what UK AI audits cost.